Can I download Qwen3.8-Max weights today?

No. Alibaba said open weights are coming 'soon' but has not shipped a public repository, model card, or license file for Qwen3.8-Max. Access today is preview-only through paid Alibaba subscription tiers — the Token Plan, Qoder, and QoderWork — at 10% of the standard token price.

How does Qwen3.8-Max compare to Kimi K3 and Claude Fable 5?

Alibaba positions Qwen3.8-Max as 'second only to Fable 5,' but no independent benchmarks have been published. Kimi K3, released three days earlier by Moonshot, has posted 76.8% on SWE-bench Verified and other scores. Qwen3.8-Max's true ranking will only be knowable once Artificial Analysis, LMArena, and Vals AI run their evaluations.

What is Qwen3.8-Max's context window and architecture?

Alibaba has not disclosed Qwen3.8-Max's context window in its July 19 announcement. Architecture details — the mixture-of-experts structure, active parameters per token, training compute, and modality-specific components — have not been shared. Only the 2.4T total parameter count and general capability positioning are public so far.

Where can I try Qwen3.8-Max right now?

Qwen3.8-Max preview access ships through Alibaba's paid tiers: Token Plan, Qoder, and QoderWork. Pricing is 10% of the standard rate for the preview window. There is no free public playground, no chatbot on qwen.ai, and no HuggingFace endpoint yet.

Alibaba (Qwen) · 2026-07-19 · major

Qwen3.8-Max Preview — Alibaba's 2.4T multimodal flagship arrives, weights promised

Alibaba unveils Qwen3.8-Max, a 2.4-trillion-parameter multimodal model. Preview access is live via the Alibaba Token Plan, Qoder, and QoderWork at 10% of standard price. Open weights are promised but not shipped.

Alibaba announces Qwen3.8-Max, a 2.4T multimodal preview, and promises open weights soon.

Quick facts

Maker	Alibaba (Qwen team)
Total parameters	2.4T (Alibaba's number)
Modalities	Text, image, video, documents
Availability	Preview on Token Plan, Qoder, QoderWork
Preview pricing	10% of standard rate
Open weights	Promised, no date set
Benchmarks	None published

What is it?

Qwen3.8-Max is Alibaba's new flagship model at 2.4 trillion parameters. The Qwen team calls it their first multimodal model above one trillion parameters, handling images, videos, and documents alongside text. Preview access opened on July 19 through the Alibaba Token Plan, Qoder, and QoderWork at 10% of the standard token price. Open weights are promised but no repository, model card, or license has shipped yet.

How does it work?

The preview runs as a paid API on Alibaba's Cloud stack. Alibaba has not disclosed the mixture-of-experts structure, active parameter count, context window, or training compute for Qwen3.8-Max. The team says the model 'should outperform Qwen 3.7-Max, especially in coding and complex productivity tasks such as full-stack development, data analysis, and office workflows.'

Why does it matter?

Qwen3.8-Max is the biggest model announcement of the week, arriving three days after Moonshot's Kimi K3. Alibaba's claim that Qwen3.8 sits 'second only to Fable 5' is unverified — no benchmark tables have been published — but a 2.4T flagship from a top Chinese lab with open weights on the roadmap will pressure closed labs and open competitors as soon as Artificial Analysis and LMArena post independent scores.

Who is it for?

developers testing frontier Chinese models on paid APIs

Frequently asked questions

Can I download Qwen3.8-Max weights today?: No. Alibaba said open weights are coming 'soon' but has not shipped a public repository, model card, or license file for Qwen3.8-Max. Access today is preview-only through paid Alibaba subscription tiers — the Token Plan, Qoder, and QoderWork — at 10% of the standard token price.
How does Qwen3.8-Max compare to Kimi K3 and Claude Fable 5?: Alibaba positions Qwen3.8-Max as 'second only to Fable 5,' but no independent benchmarks have been published. Kimi K3, released three days earlier by Moonshot, has posted 76.8% on SWE-bench Verified and other scores. Qwen3.8-Max's true ranking will only be knowable once Artificial Analysis, LMArena, and Vals AI run their evaluations.
What is Qwen3.8-Max's context window and architecture?: Alibaba has not disclosed Qwen3.8-Max's context window in its July 19 announcement. Architecture details — the mixture-of-experts structure, active parameters per token, training compute, and modality-specific components — have not been shared. Only the 2.4T total parameter count and general capability positioning are public so far.
Where can I try Qwen3.8-Max right now?: Qwen3.8-Max preview access ships through Alibaba's paid tiers: Token Plan, Qoder, and QoderWork. Pricing is 10% of the standard rate for the preview window. There is no free public playground, no chatbot on qwen.ai, and no HuggingFace endpoint yet.

Try it

Alibaba Token Plan preview (qwen.ai / Qoder / QoderWork)